Lilian Besson - professor at ENS Rennes

📜 Short Bio

Hello ! I am Lilian Besson, a former student in Mathematics and Computer Science at ENS de Cachan. I am a passionate programmer, open-source enthusiast and young researcher in machine learning, learning theory and cognitive radio. I also love to cook and meet people, to travel and exchange, to bike or hike. 👨 I use the he/him pronouns.

From August 2019, I am now a junior professor (agrégé) at ENS Rennes, in charge of the class preparing the "agrégation" national exam, with a major in mathematics and a minor in computer science, level M2, and in charge of lectures for introduction and advanced algorithms.
I am also an associate researcher with the PANAMA team at IRISA and INRIA Rennes. My research interests lie between sequential learning, tensor decomposition, cognitive radio, Python and open source software, and other directions.

Between October 2016 and November 2019:

I was pursuing my PhD at CentraleSupélec (SCEE team, IETR lab) in Rennes (France), with Christophe Moy and Émilie Kaufmann (CNRS & Inria, SequeL team, CRIStAL lab, at Lille). My PhD is on multi-players multi-arms bandits models applied to radio-telecommunication, especially Internet-of-Things problems. I defended in November, and I now hold a PhD in telecommunications. My PhD thesis is here and the slides used for my defense are there.
And I am teaching theoretical computer science at ENS Rennes (for a class preparing for the "agrégation" national exam, level M2) and ENSAI (complexity and calculabilty, level L3).

📰 News

September 2019

I have finished my contract as PhD student, I am starting as junior professor (agrégé) at École Normale Supérieure de Rennes, within the IT and Telecommunications Department. I will teach Algorithms and theoretical computer science.
June 2019

I finished writing my PhD thesis! I have sent it for the reviewers on the 12th of July, and I will defend my PhD on the 20th of November 2019!
February 2019

We finished our article "The Generalized Likelihood Ratio Test meets klUCB: an Improved Algorithm for Piece-Wise Non-Stationary Bandits", as well as a short version in French that was accepted to the GRETSI 2019 conference.
January 2019

My two articles sent to the IEEE WCNC conference were accepted! I was in Marrakech (Marrocco) in April 2019.
March, 2018

I released the 📃 documentation and the 🗄 full code of my numerical environments for Multi-Armed Bandits problems 🎰, SMPyBandits.
October, 2016

I started my Ph.D. ! I am fully funded for 3 years, with a grant from the Ministry of Higher Education and Research. Thanks to them! I'm almost done! I'll defend in November 2019!

🔬 Research

For my Ph.D., my research is in applied machine learning, focused on low-cost online learning algorithm with limited feedback (bandit feedback), mainly applied to cognitive radio problems for Opportunistic Spectrum Access and setting up reliable network access protocol for the future Internet of Things networks. By studying and applying classical and recent Multi-Armed Bandit algorithms to carefully designed radio models, we are able to prove some performance guarantees, both numerically in simulations and theoretically with statistical proofs.
I'll continue to do research after my PhD, I'm joining on next September the PANAMA team at IRISA laboratory in Rennes!

14 Research talks and posters
(since 2017)
10 Research articles
(8 published, 2 sent)
2 PhD advisors
and collaboration with 2 PhD student and postdoc
2 Higher Education institutes where I teach in Rennes
1 Research software
(and many personal projects!)

📚 Publications

Orcid.org logo Orcid • arXiv • DBLP • IdHAL • Google Scholar • HALtools
List of the PDF of my articles

Preprints or in process of submissions

C. Moy & L. Besson & G. Delbarre & L. Toutain.
Decentralized Spectrum Learning for Radio Collision Mitigation in Ultra-Dense IoT Networks: LoRaWAN Case Study and Measurements.
Accepted to a special volume of the Annals of Telecommunications journal, on "Machine Learning for Intelligent Wireless Communications and Networking", August 2020. [ PDF ] • [ DOI ]
L. Besson & E. Kaufmann.
The Generalized Likelihood Ratio Test meets klUCB: an Improved Algorithm for Piece-Wise Non-Stationary Bandits. February 2019.
[ PDF ] • [ Mobile version ] • [ Slides ] • [ arXiv ] • [ HAL ] • [ BibTeX ] • [ Code (LaTeX) ] • [ Code ]
L. Besson & E. Kaufmann.
What Doubling-Trick Can and Can't Do for Multi-Armed Bandits. September 2018.
[ PDF ] • [ Mobile version ] • [ arXiv ] • [ HAL ] • [ BibTeX ] • [ Code (LaTeX) ] • [ Code ]
L. Besson. 🔬 🎰 SMPyBandits: an Open-Source Research Framework for Single and Multi-Players Multi-Arms Bandits (MAB) Algorithms in Python. July 2018.
[ PDF ] [ PDF (long version) ] • [ HAL ] • [ BibTeX ] • [ Code ] • [ Documentation ]
L. Besson. A Note on the Ei Function and a Useful Sum-Inequality. February 2018.
[ PDF ] • [ HAL ] • [ BibTeX ] • [ Code ]

Communications in National Conferences (in France)

L. Besson & E. Kaufmann.
Analyse non asymptotique d'un test séquentiel de détection de ruptures et application aux bandits non stationnaires (in French).
GRETSI, Lille, France. August 2019.
[ PDF ] • [ 🇬🇧 Slides ] • [ 🇫🇷 Slides ] • [ HAL ] • [ BibTeX ] • [ Code (LaTeX) ] • [ Code ]

Communications in International Conferences

C. Moy & L. Besson.
Decentralized Spectrum Learning for IoT Wireless Networks Collision Mitigation.
1st ISIoT workshop at the DCOSS conference, Santorini, Greece. May 2019.
[ PDF ] • [ arXiv ] • [ HAL ] • [ BibTeX ] • [ DOI ]
R. Bonnefoi, L. Besson, J. Manco-Vasquez & C. Moy.
Upper-Confidence Bound for Channel Selection in LPWA Networks with Retransmissions.
1st MoTION workshop at WCNC, Marrakech, Morocco. April 2019.
[ PDF ] • [ Mobile version ] • [ arXiv ] • [ HAL ] • [ BibTeX ] • [ Slides ] • [ Code (LaTeX) ] • [ Code (MATLAB) ] • [ DOI ]
L. Besson, R. Bonnefoi & C. Moy.
GNU Radio Implementation of MALIN: "Multi-Armed bandits Learning for Internet-of-things Network".
WCNC (Wireless Communication and Networks Conference), Marrakech, Morocco. April 2019.
[ PDF ] • [ Mobile version ] • [ arXiv ] • [ YouTube Video ] • [ HAL ] • [ BibTeX ] • [ Slides ] • [ Code (LaTeX) ] • [ Code (GNU Radio) ] • [ DOI ]
L. Besson, R. Bonnefoi & C. Moy.
MALIN: Multi-Arm bandit Learning for Iot Networks with GRC: A TestBed Implementation and Demonstration that Learning Helps.
Demonstration presented at the ICT (International Conference on Communication), Saint-Malo, France. June 2018.
[ PDF* ] • [ Poster ] • [ YouTube Video ] • [ Code (GNU Radio) ]

More details…
The demo implements a toy wireless system, with one IoT object communicating with a gateway (or base station) and trying to find the frequency channel which is the less perturbated by a random interfering traffic, by using reinforcement learning. We use three USRP cards by National Instruments for the random traffic, the IoT object and the gateway. It is based on our CROWNCOM 2017 article. See also this document. Here is a 5-minute video I made to explain this demonstration.
L. Besson & E. Kaufmann.
Multi-Player Bandits Revisited.
ALT, Lanzarote, Canari Islands, April 2018.
[ PDF ] • [ Mobile version ] • [ Erratum ] • [ Slides ] • [ Poster ] • [ arXiv ] • [ HAL ] • [ BibTeX ] • [ Code (LaTeX) ] • [ Code ]

Presented a few times…
In 2017 and 2018, I presented different versions of these slides for a SequeL seminar in Lille on December 22th, for the ALT conference in Lanzarote (Spain) on April 8th 2018, and I presented this poster for a Workshop on Multi-Armed Bandits and Learning Algorithms on May 24th 2018 in Rotterdam School of Management (Erasmus University, Netherlands), for the IETR lab "PhD Students Day" on June 15th 2018 (Vannes, France), and one last time for the Workshop Optimization and Learning on September 10th 2018 in University de Toulouse in Toulouse (France).
L. Besson, E. Kaufmann & C. Moy.
Aggregation of Multi-Armed Bandits learning algorithms for Opportunistic Spectrum Access.
IEEE WCNC, Barcelona, Spain. April 2018.
• [ PDF ] • [ Mobile version ] • [ Slides ] • [ HAL ] • [ BibTeX ] • [ Code (LaTeX) ] • [ Code ] • [ DOI ] • [ Actes ]
R. Bonnefoi, L. Besson, C. Moy, E. Kaufmann & J. Palicot.
Multi-Armed Bandit Learning in IoT Networks and non-stationary settings.
CrownCom, Lisboa, Portugal. September 2017. Best Paper Award!
[ PDF ] • [ Mobile version ] • [ Slides ] • [ Poster ] • [ arXiv ] • [ HAL ] • [ BibTeX ] • [ Code (LaTeX) ] • [ Code (MATLAB) ] • [ DOI ] • [ Proceedings ]

Presented a few times…
In 2017 I presented different versions of these slides for a SequeL seminar in Lille on September 15th, for the CrownCom conference in Lisboa on September 22th, for this research day organized by the GdR-ISIS in Paris on November 17th, and for the SCEE seminar in Rennes on November 23th.

Academic publications

L. Besson.
Multi-players Bandit Algorithms for Internet of Things Networks.
PhD thesis. Under supervision of Prof. Christophe Moy and Dr Emilie Kaufmann. Team SCEE at CentraleSupélec (campus de Rennes) and IETR, Rennes, and team SequeL at Inria Lille Nord Europe, Lille. October 2016 - November 2019.
[ PDF ] • [ bibTeX ] • [ Code (thesis) ] • [ Slides ] • [ Code (slides) ] • [ Code (simulations) ]

Summary of my PhD thesis
In this PhD thesis, we study wireless networks and reconfigurable end-devices that can access Cognitive Radio networks, in unlicensed bands and without central control. We focus on Internet of Things networks (IoT), with the objective of extending the devices' battery life, by equipping them with low-cost but efficient machine learning algorithms, in order to let them automatically improve the efficiency of their wireless communications. We propose different models of IoT networks, and we show empirically on both numerical simulations and real-world validation the possible gain of our methods, that use Reinforcement Learning. The different network access problems are modeled as Multi-Armed Bandits (MAB), but we found that analyzing the realistic models was intractable, because proving the convergence of many IoT devices playing a collaborative game, without communication nor coordination is hard, when they all follow random active pattern. The rest of this manuscript thus studies two restricted models, first multi-players bandits in stationary problems, then non-stationary single-player bandits. We also detail another contribution, SMPyBandits, our open-source Python library for numerical MAB simulations, that covers all the studied models and more. Keywords: Internet of Things (IoT), Cognitive Radio, Learning Theory, Collision Mitigation Sequential Learning, Reinforcement Learning, Multi-Armed Bandits (MAB), Decentralized Learning, Multi-Player Multi-Armed Bandits, Change Point Detection, Non-Stationary Multi-Armed Bandits.
L. Besson, J. Fageot & M. Unser.
A Theoretical Study Of Steerable Homogeneous Operators, And Applications.
Research Internship Report – Master MVA. Internship with Prof.Dr. Michael Unser and Dr. Julien Fageot. BIG team at EPFL, Lausanne (Switzerland). August 2016.
[ PDF ] • [ Slides ] • [ Code ] • [ 🏅 Rank: 1st & 🎓 Grade: 18.43/20 ]

🎓 Teaching

My wish is to keep doing both teaching and research, in Computer Science and/or Maths. Ideally, I will be the happiest man if I could spend the next 40 years teaching introductory Computer Science with Python, data structures and algorithmics, computer architecture etc. I like to share my passion for practical computer science and mathematics, and how both can be combined and used for research and modelisation problems, for instance for analysis of data coming from our daily life (e.g., my text messages, quotes from the Kaamelott TV show, or data about the national result for the maths agrégation exam, etc).
I am still teaching after my PhD, and even much more, as I joined since September the Computer Science and Telecommunications department de at ENS de Rennes (as a junior professor), see this page.

2019-2020 @ ENS de Rennes, Rennes, France :
• I am responsible of the training for Computer Science option for the national exam of "agrégation de mathématiques" (to become a teacher in maths and/or computer science) ("prépa agrég", option D), and in particular I'm in charge of the practical sessions of programming, in Python 🐍 and OCaml 🐫. (80 h / year) The time table is here!
• I give the course of Introduction to Algorithmic (ALGO1) for the first year student (in L3) at ENS de Rennes (see the page for the same course in the maths department, given by François Schwarzentruber). (20h / year)
• I give the tutorial sessions for the course of Advanced Algorithmic (ALGO2), for the first year student (in L3) at ENS de Rennes, given by Nathalie Bertrand. (20h / year)
• I'm also in charge of the bi-monthly research seminar of the Computer Science department, of some visits of research labs, and of the internships of first year students (in L3) at ENS de Rennes. (40h / year)
2016-2019 @ ENS de Rennes, Rennes, France :
"préparation à l'agrégation", with notebooks in Python 🐍 and OCaml 🐫. (40 h / year)
2016-2018 @ ENSAI, Rennes, France :
calculability and complexity theory with Turing Machines 🖥 (24 h / year).
2016-2017 :
various teaching and scientific resources, many Jupyter notebooks, and some slides. Written regularly, covering various subjects. Lots of geeky stuff in English, but teaching resources are in French.
2015-2016 @ Lycée Lakanal, Sceaux, France :
maths oral exams and Computer Science practical sessions in Prep'School (MP) (120 h / year).
2014-
active and regular participation in the writing of a mathematics course, which covers quite completely the French mathematics agrégation ("Le Frido"), with Laurent Claessens. The sources of the book are distributed under open-source GPL license on GitHub (@LaurentClaessens/mazhe), and the PDF version of the book is freely distributed.

More details...
In addition to more than 2000 pages, the four volumes of the Frido contain mathematics at the aggregation level. The book covers almost the entire program by trying to keep the ratings consistent from one end to the other (no rating abuse that the reader should guess), while remaining clear and very complete in the demonstrations. No holes that the reader should fill himself, nothing is a priori considered obvious. The book has been published since 2016 under the title "Le Frido", currently in four volumes with an ISBN number and therefore authorized for oral aggregation.
In particular, I have contributed to the chapters on statistics, and corrected many typos. I use this book very regularly, for my mathematics lessons in pre-preparation and preparation for aggregation, and to maintain myself as much as possible at the level of aggregation. In general, I read it again from one end to the other once or twice a year.... This book remains more than 99.99% of Laurent Claessens' creation.
2014-2015 @ Mahindra École Centrale, Hyderabad, India :
full-time teaching assistant for 2 maths courses and 1 Computer Science course, MA101 & MA102, CS101 (250 h / year).
2012-2014 @ Correction Rue Ulm, Paris, France :
one-to-one teaching assistant for junior high maths (30 h / year), and grading of maths written exams (10 times / year).

💻 Code

I love to program and write code and documentation , especially in Python for science, GNU Bash for desktop automation, and HTML/CSS/JS for the web. Everything I did and do is open-source and published (using Git), on my GitHub or my Bitbucket profiles.

Expert: 🐍 Python (research & teaching), GNU Bash (scripts), Caml/OCaml (teaching).
Tools: Jupyter notebooks for OCaml and 🐍 Python, OCamlDoc for OCaml, Sphinx for 🐍 Python, databases (SQLite, SQL), version control software (git, svn, hg).
Proficient: HTML & CSS & JavaScript (web), Julia & GNU Octave & MATLAB & SciLab (numerical computations), C/C++.
Knowledgable: BASIC for Casio and Texas Instrument graphical calculator, GNU Radio, nVidia CUDA, Lua.
Computers: expert user, capable of managing dozens of machines, under Windows or GNU/Linux.
Writing: daily usage of LaTeX, Beamer (for slides) and Markdown, and web pages design.

Here are a few things I did since 2012:

2018 : ParcourSup.py 🇫🇷 A clone of the code used by French government to decide the affectation between highschool students and universities (ParcoursSup), written in 🐍 Python 3. Aimed at being didactic. Documentation here (in French), and an interactive notebook here (in French).
2018 : Gym-NES-Mario-Bros 🐍 🏋 OpenAI GYM for Nintendo NES emulator FCEUX and 1983 game Mario Bros. + Double Q Learning for mastering the game with reinforcement learning.
2017 : Jupyter-NBConvert-OCaml , custom Jupyter 📓 NBConvert exporter for the 🐫 OCaml language (and ocaml-jupyter kernel)
2017 : small Python 🐍 (Lempel-Ziv_Complexity) and Julia (LempelZiv.jl) libraries to efficiently compute the Lempel-Ziv complexity between two binary strings. Published on Pypi and METADATA.jl
2016 : uLogMe , self-monitoring software for GNU/Linux, with zero CPU overhead and outstanding visualizations.
2013 : StrapDown.js , An awesome Javascript tool to quickly publish nice-looking webpages from raw Markdown, no server-side compilation. Light script .
Since 2012 : My résumé 🎓, is open-source and publicly available, self hosted.